Analysis of cancer genomes reveals basic features of human aging and its role in cancer development

Abstract

Somatic mutations have long been implicated in aging and disease, but their impact on fitness and function is difficult to assess. Here by analysing human cancer genomes we identify mutational patterns associated with aging. Our analyses suggest that age-associated mutation load and burden double approximately every 8 years, similar to the all-cause mortality doubling time. This analysis further reveals variance in the rate of aging among different human tissues, for example, slightly accelerated aging of the reproductive system. Age-adjusted mutation load and burden correlate with the corresponding cancer incidence and precede it on average by 15 years, pointing to pre-clinical cancer development times. Behaviour of mutation load also exhibits gender differences and late-life reversals, explaining some gender-specific and late-life patterns in cancer incidence rates. Overall, this study characterizes some features of human aging and offers a mechanism for age being a risk factor for the onset of cancer.

Introduction

Recent analyses of human whole-genome germline mutations revealed that mutation load in offspring behaves on average as a monotonic function of paternal age, affecting the risk of autism and schizophrenia in the children of older fathers1,2,3,4. Growth of somatic mutation load and burden also modulates the risk of disease by increasing the likelihood of mutations directly affecting the relevant genes and perturbing gene regulatory networks. For example, studies suggest that mitochondrial DNA mutations are involved in the development of Alzheimer’s and Parkinson’s diseases5,6, and an aging-associated increase in mtDNA mutations would thus lead to an increase in the risk of these neurological diseases with age. Cancers grow from clonal expansions of single mutated somatic cells. Initiation of clonal expansion and transition from a non-malignant to malignant regime are associated with mutations in two to seven gene drivers of cancer7, and an age-dependent increase in the overall somatic mutation load naturally leads to a proportional increase in somatic mutation burden, affecting probability of driver mutations and making cancer a disease of aging8,9.

Although identification and quantitative analysis of age-related somatic mutations are extremely important for understanding the aetiology and estimating the baseline risk of cancer and other diseases of aging, such analysis remains a major technical challenge. Human somatic mutation accumulation rates were previously estimated to be within the range of 70–700 mutations per year in proliferating cells, and an order of magnitude lower in non-replicating cells10. Variability in estimated mutation rates highlights many difficulties associated with accurate measurement and quantitative analysis of age-related somatic mutations. In particular, since somatic mutations differ among individual cells of the organism11, numerous single cell genomes from subjects of different age are needed to assess the age-dependence of average mutation load and to quantify mutation accumulation rates.

A key feature of somatic mutations in proliferating cells is that they may lead to cancer. It is known that somatic mutation numbers increase with age in many cancers12,13, consistent with earlier reports of age-related accumulation of mutations in model animals11. Genomes of cells in the clone include multiple mutations originating well before the onset of cancer13, and clones may be thought of as ‘lenses’ amplifying mutational patterns in single cells observed at premalignant stages. Thus, cancer genomes become a powerful tool for characterizing age-related accumulation of human somatic mutations14, and an alternative to single cell technologies. Here we show that analyses of cancer genomes allow estimation of the aging-associated increase in mutation load and burden, in turn leading to insights into both aging and cancer.

Results

Non-linear increase and slowdown of mutation load with age

Taking advantage of the availability of thousands of human cancer genomes in the Cancer Genome Atlas (TCGA)15, we quantitatively assessed age-related mutational patterns in cancers. First, we combined different cancer types sequenced by five major research centers (Baylor College of Medicine, Broad Institute, Canada’s Michael Smith Genome Sciences Centre, the University of California Santa Cruz and Washington University School of Medicine) and analysed the resulting statistical ensembles of whole-exome samples to assess multi-tissue age-related changes in mutation load. We found that each of the ensembles produced by different sequencing centres was characterized by a non-uniform distribution of cumulative mutation load (Fig. 1a–c), with most cancer samples having <150 mutations/exome and a few samples showing much larger numbers of mutations. As estimated values of somatic mutation load in a given sample depend on the choice of sequencing technology and variant-calling pipelines, and because the lists of cancer types sequenced by individual centres varied, the median mutation loads and their behaviour with age were slightly different among the ensembles of samples produced by different centres. However, in all cases, they were located at relatively low values of 10–150 mutations per exome, and grew with age monotonically but non-linearly, significantly slowing down in late life (Fig. 1d). This non-linear accumulation of somatic mutations with age is also consistent with recent estimations of the behaviour of mutation load with age in hematopoietic clones16,17.

Figure 1: Accumulation of mutations with age and cancer incidence.

(a) Somatic mutation distribution and combined mutation load for gender-nonspecific cancers (ESCA, LIHC, PAAD, PCPG, SARC, STAD and UVM in both men and women) sequenced by Canada’s Michael Smith Genome Sciences Centre (BCGSC). Each point represents a particular cancer sample. Black line denotes behaviour of the median of the distribution with age. (b) The same for samples sequenced by Baylor College of Medicine (BCM); cancers include ESCA, LIHC, LGG, PAAD, PCPG, STAD, THCA and UVM. (c) The same for samples sequenced by the University of California-Santa Cruz (UCSC); cancers include ESCA, KIRP, LIHC, PAAD, PCPG, SARC and UVM. (d) Behaviour of the mutation load distribution medians with age for distributions depicted in a–c. (e) Mutation load distribution for patients with colon adenocarcinoma (COAD), samples sequenced by BCM, IlluminaGA pipeline. Cancers in men are shown in blue, and in women in red. (f) Dynamics of the mutation load distributionP(N, t) of somatic mutation counts N with age t for breast adenocarcinoma (BRCA) in women. Samples sequenced by Washington University, IlluminaGA human-curated pipeline. The actual histograms of mutation loads are approximated by 8 degree polynomials. For each age cohort, the distribution peak is denoted by a coloured arrow. (g) Somatic mutation load versus cancer incidence (cases per 1,000 capita) for pheochromocytoma and paraganglioma (PCPG) cancers in women; samples sequenced by Broad Institute, IlluminaGA automated variant-calling pipeline. Cancer incidences (green) and age-related somatic mutation accumulation pattern (blue). (h) The same for cervical squamous cell carcinoma and endocervical adenocarcinoma (CESC). Samples sequenced by BCGSC, IlluminaHiSeq automated pipeline. (i) The same for pancreatic adenocarcinoma (PAAD) in men. Samples sequenced by BCM, IlluminaGA automated pipeline. Results for other cancers, datacenters and variant-calling pipelines are presented in Supplementary Figs 1–16. Errors are s.d., calculated using bootstrapping as described in Methods.